PROCEEDINGS OF THE AMERICAN MATHEMATICAL SOCIETY Volume 132, Number 2, Pages 503–512 S 0002-9939(03)07063-1 Article electronically published on July 2, 2003
SEQUENTIAL PRODUCT OF QUANTUM EFFECTS AURELIAN GHEONDEA AND STANLEY GUDDER (Communicated by Joseph A. Ball)
Abstract. Unsharp quantum measurements can be modelled by means of the class E(H) of positive contractions on a Hilbert space H, in brief, quantum effects. For A, B ∈ E(H) the operation of sequential product A◦B = A1/2 BA1/2 was proposed as a model for sequential quantum measurements. We continue these investigations on sequential product and answer positively the following question: the assumption A ◦ B ≥ B implies AB = BA = B. Then we propose a geometric approach of quantum effects and their sequential product by means of contractively contained Hilbert spaces and operator ranges. This framework leads us naturally to consider lattice properties of quantum effects, sums and intersections, and to prove that the sequential product is left distributive with respect to the intersection.
1. Introduction Unsharp quantum measurements experiments can be modelled by means of the class E(H) of positive contractions on a Hilbert space H. For A, B ∈ E(H) the operation of sequential product A ◦ B = A1/2 BA1/2 was proposed as a model for sequential measurements. A careful investigation of properties of the sequential product has been carried over in [5] (see also [6]). In that paper it was proved that, if the underlying Hilbert space H is finite dimensional, then, by preceding the effect B with another effect A, we cannot amplify the effect B. In terms of the operator model, this means that if for some A, B ∈ E(H) we have A ◦ B ≥ B, then AB = BA = B. The question, raised in [5], is whether this is true for infinite-dimensional Hilbert spaces as well. We answer positively this question in Theorem 2.6. The idea of our proof is to iterate the inequality A ◦ B ≥ B and to show that, at the limit, we obtain exactly the condition AB = BA = B. Here we use the borelian functional calculus for selfadjoint operators, that is, we pass to the von Neumann algebra generated by the operator A to get at the limit the orthogonal projection onto Ker(I − A), a common procedure in ergodic theory. The second section also contains a few other results on the sequential product, especially concerning comparison. In the third section we propose a geometric approach of the sequential product based on the interpretation of quantum effects as contractively contained Hilbert spaces. This is a particular situation of the theory of continuously contained Hilbert spaces in quasi-complete locally convex spaces of L. Schwartz [8]. Similar ideas have Received by the editors August 29, 2002 and, in revised form, October 17, 2002. 2000 Mathematics Subject Classification. Primary 47B65, 81P15, 47N50, 46C07. c
2003 American Mathematical Society
503
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
504
AURELIAN GHEONDEA AND STANLEY GUDDER
been used in the case of contractively contained Hilbert spaces in the de BrangesRovnyak model in quantum scattering [2], and operator ranges; cf. [4] and [7]. This allows us to consider lattice properties, as sums and intersections of the set of quantum effects, that lead naturally to the parallel sum of quantum effects. The sequential product is a morphism, in the second variable, with respect to the binary operation of parallel sum; cf. Theorem 3.10. Rephrased in terms of contractively contained Hilbert spaces, this means that the sequential product is left distributive with respect to the intersection. 2. Sequential product Let H be a Hilbert space. We denote by B(H) the C ∗ -algebra of linear and bounded operators on H, B(H)+ = {A ∈ B(H) | A ≥ 0} its positive cone, and by E(H) we denote the set of quantum effects, or simply, effects, that is, operators A ∈ B(H) such that 0 ≤ A ≤ I. For two effects A, B ∈ E(H) we denote by A ◦ B = A1/2 BA1/2 the sequential product of the effects A and B, that is, the effect obtained by doing the measurements in the order first A and second B. Clearly, A ◦ B ≤ A and a first question is when A ◦ B ≤ B. A sufficient condition to have this is the case when A and B are compatible, that is, AB = BA, but we would like to have a complete characterization. The answer is given by a particular case of the general majorization theorem (e.g. see [3]). Theorem 2.1. Let Ti ∈ B(Hi , G) for Hilbert spaces Hi and G, i = 1, 2. Then for 0 ≤ γ < ∞ the following conditions are equivalent: (a) kT1∗ gkH1 ≤ γkT2∗ gkH2 for all g ∈ G; (b) T1 T1∗ ≤ γT2 T2∗ ; (c) T1 = T2 S for some operator S ∈ B(H1 , H2 ) with kSk ≤ γ. Each of these equivalent conditions implies (d) Ran(T1 ) ⊆ Ran(T2 ). Conversely, (d) implies all of (a) through (c), for some γ < ∞, the operator S can be chosen such that Ker(T1 ) ⊆ Ker(S) and Ker(T2 ) ⊆ Ker(S ∗ ), in which case it is uniquely determined. As a consequence, given two operators X, Y ∈ B(H), we have XX ∗ ≤ Y Y ∗ if and only if X = Y T for some T ∈ B(H), kT k ≤ 1. In the following, for A ∈ E(H) we denote by PA the selfadjoint projection onto the closure of Ran(A) (which is the same with the closure of Ran(A1/2 )). Theorem 2.2. Let A, B ∈ E(H) be effects. Then A ≤ B if and only if there exists another effect C ∈ E(H) such that A = B ◦ C. In addition, C can be chosen such that C ≤ PA and in this case it is uniquely determined. Proof. Clearly, if A = B ◦ C for some effect C ∈ E(H), then A = B ◦ C ≤ B. Conversely, let us assume that A ≤ B. Then A1/2 A1/2 ≤ B 1/2 B 1/2 and hence, by Theorem 2.1, there exists T ∈ B(H), such that kT k ≤ 1 and A1/2 = B 1/2 T . Then C = T T ∗ ∈ E(H) and hence A = B 1/2 T T ∗B 1/2 = B ◦ C. The uniqueness part follows from the uniqueness part of Theorem 2.1. Corollary 2.3. Let A, B ∈ E(H). Then A ◦ B ≤ B if and only if there exists C ∈ E(H) such that A ◦ B = B ◦ C.
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
SEQUENTIAL PRODUCT OF QUANTUM EFFECTS
505
Corollary 2.3 says that the effect A ◦ B diminishes the effect B if and only if there is another effect C such that, performing first the measurement B and then C, the obtained sequential measurement B ◦ C is the same as A ◦ B. In Theorem 3.2 in [5] it is proved that, given A, B ∈ E(H), we have the associativity property (A ◦ B) ◦ C = A ◦ (B ◦ C), for all C ∈ E(H), if and only if A and B are compatible, that is, AB = BA. As a consequence of Theorem 2.2, we have: Corollary 2.4. If A, B, C ∈ E(H) are effects, then there exists an effect D ∈ E(H) such that A ◦ (B ◦ C) = (A ◦ B) ◦ D. Proof. Since B ◦ C ≤ B, we have A ◦ (B ◦ C) ≤ A ◦ B. Then by Theorem 2.2, there exists D ∈ E(H) such that A ◦ (B ◦ C) = (A ◦ B) ◦ D. For a fixed A ∈ E(H), the mapping E(H) 3 B 7→ sA (B) = A ◦ B ∈ E(H) is strongly continuous and affine. We are interested in sections of the sequential mapping sA . Since A ◦ B ≤ A, it follows that such a section should be defined on [0, A] = {C ∈ E(H) | C ≤ A}. Recall that PA denotes the selfadjoint projection onto the closure of Ran(A). By Theorem 2.2, for any B ∈ [0, A] there exists a unique C ∈ [0, PA ] such that B = A ◦ C. Denote B/A = C and call it the sequential quotient of B by A. In this way, we can define a mapping (2.1)
fA : [0, A] → [0, PA ],
fA (B) = B/A, B ∈ [0, A].
Clearly, fA is a section of sA . Let us also note that the “segment” [0, A] has a natural sequential product, that we denote by ×, (2.2) C × D = A ◦ (C/A) ◦ (D/A) , C, D ∈ [0, A]. Theorem 2.5. For any A ∈ E(H), the mapping fA defined as in (2.1) is an affine strongly continous homeomorphism, such that (2.3)
fA (C × D) = fA (C) ◦ fA (D),
C, D ∈ [0, A].
Proof. Note first that fA is bijective: its inverse is sA |[0, PA ] : [0, PA ] → [0, A]. Thus, since sA is affine and [0, A] is convex, sA |[0, PA ] is affine, hence fA is affine. To see that fA is strongly continuous, note that hBx, yi = h(B/A)A1/2 x, A1/2 yi,
B ∈ E(H), x, y ∈ H,
and recall that the strong operator topology coincides with the weak operator topology on bounded sets of positive operators. Thus, fA is an affine strongly continuous homeomorphism. The property expressed in (2.3) is just another way of writing (2.2). We are now interested in the question of when A ◦ B ≥ B. In Theorem 2.6 of [5] it is proved that, if H is finite dimensional, by preceding the effect B with another effect A, we cannot amplify the effect B; more precisely, if A ◦ B ≥ B, then AB = BA = B, and it is asked whether this holds for infinite-dimensional spaces H. In the following theorem we answer this question positively. Theorem 2.6. Let A, B ∈ E(H) such that A ◦ B ≥ B. Then AB = BA = B and, consequently, A ◦ B = B.
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
506
AURELIAN GHEONDEA AND STANLEY GUDDER
Proof. In the following we will repeatedly use the following elementary fact: if C, D ∈ B(H) are selfadjoint such that C ≤ D, then for any X ∈ B(H) we have X ∗ CX ≤ X ∗ DX. Since B ≤ I, it follows that A1/2 BA1/2 ≤ A1/2 A1/2 = A. Therefore, by hypothesis we have 0 ≤ B ≤ A1/2 BA1/2 ≤ A ≤ I.
(2.4)
By applying A1/2 · A1/2 to (2.4), it follows that 0 ≤ A1/2 BA1/2 ≤ A1/2 A1/2 BA1/2 A1/2 = ABA ≤ A1/2 AA1/2 = A2 ≤ A, and hence, using again (2.4), we have 0 ≤ B ≤ A1/2 BA1/2 ≤ A2/2 BA2/2 ≤ A2 ≤ A ≤ I. Performing a similar procedure, we obtain inductively that for all n ≥ 1 we have (2.5)
0 ≤ B ≤ A1/2 BA1/2 ≤ · · · ≤ An/2 BAn/2 ≤ An ≤ · · · ≤ A ≤ I.
From (2.5) we keep only 0 ≤ B ≤ An ≤ I,
(2.6)
∀n ≥ 1.
Let us now consider the sequence of functions fn (t) = tn , fn : [0, 1] → R+ . Note that |fn (t)| ≤ 1 for all n ≥ 1 and that 1, t = 1, ∀t ∈ [0, 1]; lim fn (t) = χ{1} (t) = 0, 0 ≤ t < 1, n→∞ that is, the sequence of functions tn converges boundedly pointwise, on the compact interval [0, 1] ⊇ σ(A), to the characteristic function of the set {1}. Thus, by borelian functional calculus for selfadjoint operators we have (2.7)
so- lim An = so- lim fn (A) = χ{1} (A) = PKer(I−A) . n→∞
n→∞
Consequently, letting n → ∞ in (2.6), by (2.7) we obtain 0 ≤ B ≤ PKer(I−A) .
(2.8)
If we represent the selfadjoint operators A and B with respect to the decomposition H = Ker(I − A) ⊕ (H Ker(I − A)), by (2.8) we get
A=
I 0
0 A22
,
B=
B11 0
0 0
.
Now it is easy to see that AB = BA = B.
3. Contractively contained Hilbert spaces Let K and H be Hilbert spaces. We say that K is continuously contained in H, and we write K ,→ H, if K ⊆ H and the inclusion mapping ιK : K ,→ H is continuous, that is, there exists γ ≥ 0 such that kkkH ≤ γkkkK, for all k ∈ K. K is contractively contained in H if K ⊆ H and the inclusion mapping ιK : K ,→ H is contractive, that is, kkkH ≤ kkkK, for all k ∈ K. Let K ,→ H and denote A = ιK ι∗K ∈ B(H). Then A ≥ 0 and, following L. Schwartz [8], we call it the kernel of K relative to H.
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
SEQUENTIAL PRODUCT OF QUANTUM EFFECTS
507
In the following we propose a geometric approach to quantum effects and their sequential product. Let A ∈ B(H), A ≥ 0. Define RA = Ran(A1/2 ) and the inner product (3.1)
hA1/2 x, A1/2 yiRA = hPH Ker(A) x, yiH ,
x, y ∈ H.
More precisely, we define a Hilbert space structure on Ran(A1/2 ) in such a way that the operator A1/2 : H → Ran(A1/2 ) is a co-isometry. Thus, (RA , h·, ·iRA ) is a Hilbert space. We note that kA1/2 xkRA = kPH Ker(A) xk,
x ∈ H.
Then kA1/2 xkH = kA1/2 PH Ker(A) xkH ≤ kA1/2 k · kPH Ker(A) xk = kA1/2 k · kA1/2 xkRA ,
x ∈ H.
Thus, RA ,→ H and kιRA k ≤ kA1/2 k; in particular, if A ≤ I, then RA ,→ H contractively. Let x, y ∈ H be arbitrary. Then hιRA y, A1/2 xiRA = hy, A1/2 xiH = hA1/2 y, xiH = hPH Ker(A) A1/2 y, xiH = hA1/2 A1/2 y, A1/2 xiRA , that is, ιRA y = A1/2 A1/2 y = PH Ker(A) Ay,
y ∈ H,
ιRA ι∗RA
= A. Thus A is the kernel of RA relative to H. In particular, if and hence RA ,→ H contractively, then A ≤ I. Let K be another Hilbert space continuously contained in H and such that ιK ι∗K = A. Then Ran(A) = Ran(ι∗K ) is dense in K. For arbitray x, y ∈ H we have hAx, AyiRA = hA1/2 x, A1/2 yiH = hAx, yiH = hιK ι∗K x, yiH = hι∗K x, ι∗K yiK = hAx, AyiK ,
that is, h·, ·iRA coincide with h·, ·iK on a dense linear manifold, hence RA = K as Hilbert spaces. Thus we have proved the following theorem, which is a particular case of results in [8] (see also [2] and [4]). Theorem 3.1. Let A ∈ B(H)+ . Then there exists uniquely a Hilbert space RA ,→ H such that A is the kernel of RA relative to H. In addition, RA ,→ H contractively if and only if A ∈ E(H). Recall that for a Hilbert space H we denote E(H) = {A ∈ B(H) | 0 ≤ A ≤ I}, identified with the set of quantum effects on H, and by P(H) = {P ∈ B(H) | P = P ∗ = P 2 } the set of orthogonal projections, identified with the set of sharp effects. As a consequence of Theorem 3.1, it follows that the mapping A 7→ RA induces a bijection between E(H) and the class of Hilbert spaces contractively contained in H. This extends the bijective correspondence of P(H) with the class of subspaces of H; a Hilbert space K contractively contained in H is a subspace of H if and only if the inclusion is isometric, and in this case ιK ι∗K is the orthogonal projection of H onto K. Note that, in this correspondence, subspace means that h·, ·iK coincides with the restriction of h·, ·iH ; if we impose only the condition that the strong topology
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
508
AURELIAN GHEONDEA AND STANLEY GUDDER
of K is inherited from H, we get the class of kernels with closed range. We make this precise in our context. Corollary 3.2. Let A, B ∈ B(H)+ . The following assertions are equivalent: (i) RA ⊆ RB (set inclusion). (ii) RA ,→ RB . (iii) There exists γ ≥ 0 such that A ≤ γB. In addition, denoting ι : RA ,→ RB , the optimal constant γ in (iii) is kιk. In particular, RA ,→ RB contractively if and only if A ≤ B. Proof. Consequence of Theorem 2.1 and Theorem 3.1.
Corollary 3.3. Let A, B ∈ B(H) . The following assertions are equivalent: (i) RA = RB (equality of sets). (ii) RA = RB (equality of sets) and the norms are equivalent. (iii) There exists γ > 0 such that γ1 B ≤ A ≤ γB. +
As a consequence of Corollary 3.3, two Hilbert spaces H1 and H2 , continuously contained in the Hilbert space H, coincide as sets if and only if they coincide topologically, that is, their strong topologies coincide. On the grounds of Theorem 3.1 we can define an operation of sequential product for contractively contained Hilbert spaces. Let H1 and H2 be two Hilbert spaces contractively contained in H, and let A, B ∈ E(H) be their kernels, that is, H1 = RA and H2 = RB . Then A ◦ B = A1/2 BA1/2 ∈ E(H) and we define H1 ◦ H2 = RA◦B ,→ H contractively. Of course, this definition can be extended for continuously contained Hilbert spaces, but in this more general case we have only that H1 ◦ H2 is continuosly contained in H, unless H1 is contractively contained in H. The next corollary says that the sequential product is natural for the study of contractively contained Hilbert spaces (compare with Theorem 2.2). Corollary 3.4. Let H1 ,→ H contractively. Then, for any Hilbert space H2 ,→ H1 contractively, there exists H3 ,→ H contractively, such that H2 = H1 ◦ H3 . Proof. Consequence of Theorem 2.2 and Corollary 3.2.
Another consequence of Theorem 3.1 is that we can define a “sequential quotient” for contractively contained Hilbert spaces as the “inverse” of the operation in Corollary 3.4: if the Hilbert space H2 ,→ H1 contractively, then H2 /H1 is the unique Hilbert space H3 ,→ H contractively, such that H2 = H1 ◦ H3 . Now let K be a Hilbert space contractively contained in H. For any Hilbert spaces H1 and H2 contractively contained in K, a natural “sequential product” denoted by × can be defined as (3.2)
H1 × H2 = K ◦ ((H1 /K) ◦ (H2 /K)),
by analogy with (2.2). Then the sequential quotient is left distributive with respect to these sequential products, which is just another way of writing (3.2), by analogy with (2.3). Corollary 3.5. Let K be a Hilbert space contractively contained in H. For any Hilbert spaces H1 and H2 contractively contained in K, we have (H1 × H2 )/K = (H1 /K) ◦ (H2 /K).
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
SEQUENTIAL PRODUCT OF QUANTUM EFFECTS
509
The geometric approach of quantum effects allows us to consider the lattice properties of E(H), especially those in connection with “addition” and “intersection” of continuously contained Hilbert spaces. The addition of continously contained Hilbert spaces should correspond to addition of the kernels; cf. [8]. Here is a direct argument; cf. [4]. Let A, B ∈ B(H)+ be 1 = RA and H2 = RB . We consider the such that H bounded operator T = A1/2 −B 1/2 ∈ B(H ⊕ H, H). Then RA + RB = Ran(A1/2 ) + Ran(B 1/2 ) = Ran(T ) = Ran((T T ∗ )1/2 ) = Ran((A + B)1/2 ) = RA+B , where we took into account that, by polar decomposition, T = |T ∗ |V , where |T ∗ | = (T T ∗)1/2 and V is a partial isometry. Consequently, we have Theorem 3.6 (e.g. see [8] and [4]). Let H1 and H2 be two Hilbert spaces continuously contained in H. Then the algebraic sum H1 + H2 can be naturally organized as a Hilbert space continuously contained in H; more precisely, if A, B ∈ B(H)+ are the kernels of H1 and respectively H2 , that is, H1 = RA and H2 = RB , then RA + RB = RA+B as sets. The intersection of continuously contained Hilbert spaces is a much more subtle operation that requires the definition of parallel sum of kernels, as introduced for the finite-dimenensional case in [1], and in general in [4] (cf. [7] for a further study). Let A, B ∈ B(H)+ . Since A ≤ A + B, by Theorem 2.1 there exists a unique contraction X ∈ B(H) such that (3.3)
A1/2 = (A + B)1/2 X,
Ker(X ∗ ) ⊇ Ker(A + B).
Similarly, since B ≤ A + B, there exists a unique contraction Y ∈ B(H) such that (3.4)
B 1/2 = (A + B)1/2 Y,
Ker(Y ∗ ) ⊇ Ker(A + B).
The parallel sum of A and B is defined as (3.5)
A : B = A1/2 X ∗ Y B 1/2 .
Theorem 3.7 ([4] and [7])). Let A, B ∈ B(H)+ . Then: (i) 0 ≤ A : B ≤ A, B. (ii) A : B = B : A. (iii) Ran((A : B)1/2 ) = Ran(A1/2 ) ∩ Ran(B 1/2 ). (iv) If A1 , B1 ∈ B(H)+ are such that A ≤ A1 and B ≤ B1 , then A : B ≤ A1 : B1 . (v) If A, B 6= 0, then kA : Bk ≤ (kAk−1 + kBk−1 )−1 . (vi) h(A : B)h, hi = inf{hAa, ai + hBb, bi | h = a + b}, for all h ∈ H. (vii) If An & A and Bn & B strongly, then An : Bn & A : B strongly. According to (iii) we can give an interpretation of the binary operation of “intersection” for continuously contained Hilbert spaces as follows: Theorem 3.8. Let H1 and H2 be two Hilbert spaces continuously contained in the Hilbert space H. Then the vector space H1 ∩ H2 can be naturally organized as a Hilbert space continuously contained in H; more precisely, letting H1 = RA and H2 = R2 for A, B ∈ B(H)+ , identify H1 ∩ H2 with RA:B . If either of Hi , i = 1, 2, is contractively contained in H, that is, its kernel is a quantum effect, then H1 ∩ H2 is contractively contained in H.
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
510
AURELIAN GHEONDEA AND STANLEY GUDDER
Another simple consequence of Theorem 3.7 is the stability of the set of quantum effects under the operation of parallel sum, even with the factor 2. Corollary 3.9. Let A, B ∈ E(H). Then 2(A : B) ∈ E(H). Proof. By Theorem 3.7(i), we have A : B ≥ 0, so it remains to prove that 2(A : B) ≤ I. To see this we use Theorem 3.7(v) and the inequality between the harmonic mean and the geometric mean to see that p 2kAk kBk ≤ kAk kBk ≤ 1, k2(A : B)k ≤ kAk + kBk and hence 2(A : B) ≤ I.
We end by proving that the sequential product is a morphism, in the second variable, with respect to the parallel sum. Recall that for any A ∈ E(H) we denote [0, A] = {B | 0 ≤ B ≤ A} and PA is the orthogonal projection onto the closure of Ran(A). Theorem 3.10. Let A ∈ E(H) and consider the mapping sA : [0, PA ] → [0, A], sA (B) = A ◦ B = A1/2 BA1/2 , for all B ∈ [0, PA ]. Then sA (C : D) = sA (C) : sA (D),
C, D ∈ [0, PA ].
Proof. Let C, D ∈ [0, PA ] and consider the operators X, Y uniquely determined such that C 1/2 = (C + D)1/2 X, Ker(C ∗ ) ⊇ Ker(C + D), D1/2 = (C + D)1/2 Y,
Ker(D∗ ) ⊇ Ker(C + D).
Then, by definition, C : D = C 1/2 X ∗ Y D1/2 ,
sA (C : D) = A1/2 C 1/2 X ∗ Y D1/2 A1/2 .
Since sA (C) + sA (D) = A1/2 (C + D)A1/2 , it follows that there exists a uniquely determined partial isometry V , with appropriate supports, such that (sA (C) + sA (D))1/2 V = A1/2 (C + D)1/2 . Similarly, there exist uniquely determined partial isometries U and W , with appropriate supports, such that sA (C)1/2 = A1/2 C 1/2 U,
sA (D)1/2 = A1/2 D1/2 W.
Therefore, sA (C)1/2 = A1/2 C 1/2 U = A1/2 (C + D)1/2 XU = (sA (C) + sA (D))1/2 V XU, sA (D)1/2 = A1/2 D1/2 U = A1/2 (C + D)1/2 XW = (sA (C) + sA (D))1/2 V XW, and hence, by the definition of the parallel sum and taking into account that since C and D are in [0, PA ], it follows that multiplication on the left with A1/2 does not affect the kernels, we have sA (C) : sA (D) = sA (C)1/2 U ∗ X ∗ V ∗ V Y W ∗ sA (D)1/2 = A1/2 C 1/2 X ∗ Y D1/2 A1/2 = sA (C : D).
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
SEQUENTIAL PRODUCT OF QUANTUM EFFECTS
511
With the definition of “intersection” of continuously contained Hilbert spaces, as in Theorem 3.8, and the definition of “sequential product” for contractively contained Hilbert spaces as a consequence of Theorem 3.1, Theorem 3.10 can be rephrased to say that the sequential product is left distributive with respect to the intersection. Theorem 3.11. Let the Hilbert space H0 be contractively contained in the Hilbert space H. Then, for any Hilbert spaces K and G, contractively contained in H0 , the closure of H0 in H, we have H0 ◦ (K ∩ G) = (H0 ◦ K) ∩ (H0 ◦ G). Remark 3.12. Alternatively, Theorem 3.10 can be proved by Theorem 3.7(vi). Thus, for any quantum effects 0 ≤ C, D ≤ PA and h ∈ H we have hsA (C : D)h, hi = inf{hCx, xi + hDy, yi | A1/2 h = x + y}. Both C and D have supports in PA H = Ran(A1/2 ) and hence hsA (C : D))h, hi = inf{hCA1/2 c, A1/2 ci + hDA1/2 d, A1/2 di | A1/2 h = A1/2 c + A1/2 d} = inf{hA1/2 CA1/2 c, ci + hA1/2 DA1/2 d, di | h = c + d} = hsA (C) : sA (D)h, hi, where we take into account that, without restricting the generality, we can take all the vectors h, c, and d in PA H and that on this subspace A1/2 is one-to-one. Corollary 3.13. Let A ∈ E(H) and consider the function fA as defined in (2.1). Then, fA (E : F ) = fA (E) : fA (F ), E, F ∈ [0, A]. Proof. The function fA is the inverse of the function sA as in Theorem 3.10.
This corollary can be rephrased, in terms of the “sequential quotient” as defined in (3.2), saying that the “sequential product” of contractively contained Hilbert spaces is right distributive with respect to “intersection”. Corollary 3.14. Let K be a Hilbert space contractively contained in H. For any Hilbert spaces H1 and H2 contractively contained in K, we have (H1 ∩ H2 )/K = (H1 /K) ∩ (H2 /K). References [1] W. N. Anderson, Jr. and R. J. Duffin: Series and parallel addition of matrices, J. Math. Anal. Appl., 26 (1969), 576–594. MR 39:3904 [2] L. de Branges and J. Rovnyak: Canonical models in quantum scattering theory, in Perturbation Theory and its Applications in Quantum Mechanics (Proc. Adv. Sem. Math. Res. Center, U. S. Army, Theoret. Chem. Inst., Univ. of Wisconsin, Madison, Wis., 1965) pp. 295–392, Wiley, New York, 1966. MR 39:6109 [3] R. G. Douglas: On majorization, factorization, and range inclusion of operators on Hilbert space, Proc. Amer. Math. Soc., 17 (1966), 413–415. MR 34:3315 [4] P. A. Fillmore and J. P. Williams: On operator ranges, Advances in Math., 7 (1971), 254–281. MR 45:2518 [5] S. Gudder and G. Nagy: Sequential quantum measurements, J. Math. Phys., 42 (2001), 5212–5222. MR 2002h:81032
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
512
AURELIAN GHEONDEA AND STANLEY GUDDER
[6] S. Gudder and G. Nagy: Sequentially independent effects, Proc. Amer. Math. Soc., 130 (2002), 1125–1130. MR 2002i:81014 [7] E. L. Pekarev and Yu. L. Shmulyan: Parallel addition and parallel subtraction of operators [Russian], Izv. Akad. Nauk SSSR, Ser. Mat., 40 (1976), 366–387. MR 53:14178 [8] L. Schwartz: Sous-espaces Hilbertiens d’espaces vectoriel topologiques et noyaux associ´ es (noyaux reproduisants), J. Analyse Math., 13 (1964), 115–256. MR 31:3835 ˘ al Academiei Roma ˆ ne, C.P. 1-764, 014700 Bucures¸ti, Institutul de Matematica ˆ nia Roma E-mail address:
[email protected] Current address: Department of Mathematics, Bilkent University, 06533 Ankara, Turkey Department of Mathematics, University of Denver, Denver, Colorado 80208 E-mail address:
[email protected]
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
